The craftsmanship paradox

In 1764, James Hargreaves invented the spinning jenny. It could do the work of eight spinners simultaneously. The handloom weavers of Lancashire looked at the machine and saw extinction. They smashed a few of them, just to be safe. Within a generation, textiles were cheaper than they had ever been.

But something strange happened alongside the collapse in price. The market for hand-woven fabric didn’t just survive. It split. One part became industrial commodity. The other became luxury. And the luxury part, freed from the need to compete on volume, became more expensive and more sought-after than it had been before the machines arrived.

This is a pattern that repeats with such regularity across history that it probably deserves a name. Call it the craftsmanship paradox: every wave of automation increases the premium on the thing it can’t automate.

The pattern hides in plain sight

Photography was supposed to kill painting. When Daguerre unveiled the daguerreotype in 1839, the painter Paul Delaroche reportedly declared “from today, painting is dead.” He was wrong in the most interesting way possible. Photography didn’t kill painting. It liberated it. Painting no longer had to serve the function of accurate representation. Within decades, the Impressionists were doing things with paint that no camera could replicate. By the mid-twentieth century, a Rothko canvas (rectangles of color, no representational content at all) was selling for more than the most technically perfect photograph.

The camera automated representation. So representation stopped being the point.

Digital music tells the same story with a different instrument. When Pro Tools and home recording equipment made it trivial to produce studio-quality recordings, the value of a polished recording dropped. But live performance, the one thing you couldn’t digitize and distribute for free, went the other direction. Concert ticket revenue in North America grew from $1.5 billion in 1999 to over $6.6 billion by 2023. The easier it became to produce recorded music, the more people were willing to pay for the irreproducible experience of hearing it performed.

Desktop publishing was supposed to kill graphic design. Instead, the flood of amateur newsletters and terrible clip-art flyers made professional design more obvious and more commercially valuable. When everyone has the tools, the gap between someone who knows how to use them and someone who has taste becomes the only gap that matters.

3D printing was supposed to kill traditional manufacturing. In practice, mass-market 3D printing produced a wave of cheap plastic objects. Meanwhile, handcrafted goods entered a renaissance. Etsy went from a niche marketplace to a public company. The premium for “made by hand” grew precisely because machine-made became the default.

The paradox works like this: when machines handle the easy version, the hard version stops competing with the easy version entirely. It moves to a different market, serves a different need, and commands a different price.

Why it’s counterintuitive

The obvious prediction is always that automation destroys the manual version. And for commodities, it does. Nobody hand-spins cotton for t-shirts anymore. Nobody hand-sets type for newspapers. The commodity layer gets automated, and the people who did commodity work move on or move up.

But the prediction fails because it treats everything as a commodity. It assumes the only reason anyone did things by hand was because machines didn’t exist yet. It misses the part where some things done by hand carry meaning, context, and judgment that the machine version lacks.

A handwritten letter and a printed letter contain the same words. But they don’t mean the same thing. The handwritten one carries evidence of time, attention, and intention. The printed one is efficient. These are different categories, and automation doesn’t make the handwritten version obsolete --- it makes it rarer, which makes it more meaningful.

This distinction between commodity and craft is what most predictions about automation get wrong. They see the surface --- “a machine can now do X” --- and conclude that humans doing X are finished. But they don’t ask what X meant when a human did it, versus what it means when a machine does it.

When a human bakes bread, the bread carries the baker’s judgment: how the dough felt that day, the decision to let it rise longer, the choice to pull it a minute early. When a machine bakes bread, the bread is consistent, reliable, and interchangeable. Both are bread. They serve different purposes in the world.

The current AI moment

Software development is now entering the same split. AI can generate code, write tests, scaffold applications, and produce documentation. It does these things faster and cheaper every month. The commodity layer of software development --- the translation of clear specifications into working code --- is being automated.

And the industry’s first reaction is the same as the Lancashire weavers: part panic, part denial, part machine-smashing. Developers worry that AI will replace them. Bootcamps wonder if they’re training people for jobs that won’t exist. The discourse is full of “learn to prompt” takes that assume the future belongs to whoever can best instruct the machine.

But the pattern suggests something different. The automation of code production should increase the premium on what AI can’t produce: architectural judgment, product intuition, the ability to look at a system and know which parts are load-bearing and which are dead weight.

Think about what AI actually automates. It automates the production of code: the typing, the syntax, the boilerplate, the implementation of patterns it’s seen before. It does not automate the decision about what to build. It does not automate the ability to sit in a room with a frustrated user and understand that their real problem is not the one they described. It does not automate the taste required to look at a working feature and decide to kill it because it makes the product worse.

AI automates code production. It does not automate judgment. And the gap between production and judgment is about to become the defining axis of the industry.

The parallel to photography is almost exact. Photography automated the production of visual representation. Painting responded by doing things that weren’t about production at all. Things that were about feeling, interpretation, and a human point of view. The painters who thrived after photography were the ones who had always been doing something that a camera couldn’t capture.

What “craft” means when machines produce

Here’s where it gets interesting. “Craft” doesn’t mean doing things the old way out of nostalgia. The hand-loom weavers who thrived after the spinning jenny weren’t Luddites. They were artisans who used the machine for the parts it did well and applied their skill where the machine fell short. The best modern woodworkers use CNC routers for precision cuts and hand tools for the parts that need human judgment. The tool doesn’t diminish the craft. It reveals what the craft actually is.

For software, this means the craft was never really about writing code. Code was the medium, not the message. The craft was always about making systems that serve people well. About understanding trade-offs that don’t show up in benchmarks. About knowing that the technically correct architecture and the right architecture are often different things, because the right architecture accounts for the team that will maintain it, the business that will evolve around it, and the users who will never read the documentation.

When AI handles the mechanical production of code, what remains is the judgment layer: What should we build? For whom? What trade-offs are we willing to make? What do we intentionally leave out? How do we structure this system so it can evolve in ways we can’t predict?

These questions require context that no model has. They require understanding the politics of the organization, the unspoken constraints of the market, the difference between what users ask for and what they need. They require the kind of knowledge that only comes from shipping things, watching them break, and understanding why.

This is craftsmanship. Not code. Not syntax. Not implementation speed. The ability to make good decisions in the presence of ambiguity, and to build systems that reflect those decisions in every layer.

What to bet on

If the pattern holds (and it has held across textiles, painting, music, publishing, manufacturing, and every other domain that has gone through an automation wave) then the next decade in software will look something like this:

The commodity layer will collapse. Building a standard CRUD application, a marketing website, or a data pipeline will become trivially cheap. The people and companies that charged a premium for these outputs will face brutal price compression.

But the craft layer will expand. The demand for people who can think clearly about complex systems, who can make judgment calls under uncertainty, who can look at a product and know what’s missing. That demand will grow. And the premium they command will grow with it, because the flood of cheap commodity software will make the difference between “built” and “well-built” more visible than ever.

The split will be uncomfortable. It will devalue skills that were hard-won and genuinely useful for decades. Many people built good careers on the mechanical ability to produce software, and those careers will face the same disruption that studio musicians faced when synthesizers arrived or that typesetters faced when desktop publishing took over.

But the paradox also creates opportunity. If you’re someone who was always frustrated by the gap between what you could envision and what you could produce, if the bottleneck was always implementation and not imagination, then the machines just removed your bottleneck. You can now operate at the speed of your judgment, not the speed of your typing.

The craftsmanship paradox isn’t a threat. It’s a filter. It separates the people who were doing the hard thing all along from the people who were doing the visible thing and calling it hard.

The loom didn’t kill the weaver. The camera didn’t kill the painter. The synthesizer didn’t kill the musician. In each case, the machine killed the commodity version and elevated the craft version.

AI won’t kill the software developer. It will reveal which ones were craftspeople all along.